Please use this identifier to cite or link to this item:
Title: An Intelligent System for Online Adaptive Learning Environment
Authors: Adnan, Muhammad
Keywords: Computer & IT
Computer Science
Issue Date: 2021
Publisher: Kohat University of Science & Technology, Kohat
Abstract: Machine learning and deep learning algorithms work well where future estimations and predictions are required. Notably, in educational institutions, machine learning, and deep learning algorithms can help instructors predict learners’ learning performance. Furthermore, the prediction of learners’ learning performance can assist instructors and intelligent learning systems in taking preemptive measures/early engagement/early in tervention measures so that the learning performance of weak learners could be in creased, thus reducing learners’ failures/dropout rates. In this study, we propose an intelligent learning system that provides adaptive feed back and support to learners. The main component of an intelligent learning system is the mobile -learning (m-learning) model. Different m-learning models are created for diverse learners according to their learning features (study behavior) and feature weights. Before creating the m-learning model, learning features are elicited from learners’ mobile learning activities and stored on persistent online cloud storage. For a broader understanding of learners’ learning behavior, learning features were divided into three categories: context features, behavioral features, and final performance fea tures. Context features manifest how the learner will use mobile devices as a learning tool in different contexts. Context features include knowledge background, learning time, learning time duration, places visited, learning content type (video, text), learn ing performance, etc. Behavioral features illustrate how the learner interacts with the mobile device during the learning process. Examples of behavioral features include online learner participation in discussion groups, the number of times problems posted online, the number of times problem solved, topic repetition rate, and the number of times attempted the quiz. Lastly, the final performance feature explains the learners’ performance in the final learning activity. The final performance feature is the depen dent feature, and context/behavioral features are the independent features. In this study, we are interested in revealing how each independent feature contributes to increas ing a learner’s final performance. Knowing independent feature importance/weights in improving final performance allows intelligent learning systems/instructors to provide timely adaptive help, support, and feedback. Moreover, an intelligent learning system ii could take early engagement/early intervention measures to reduce dropout or failure rates. After features analysis and preprocessing steps, we created an m-learning model to categorize learners into performance groups (grades). Because of the high accuracy of deep learning algorithms, we used Artificial Neural Network (ANN) to classify learners into five performance groups, whereas Random Forest (RF) algorithm was used to de termine each feature’s importance and weight in the creation of the m-learning model. ANN-based m-learning model is dynamic, which means that once fully trained, the model can predict new learner performance grades given his/her learning features. To sieve the m-learning model from overfitting and underfitting problems, we trained it on learners’ features instances iteratively until we gained maximum accuracy for learners’ grades prediction. For the training ANN-based m-learning model, we used techniques such as forward propagation, back-propagation, ReLU, and Softmax activation func tion. We used the stochastic gradient descent technique for minimum error/cost in the model. It is more efficient and impressive when dealing with high dimensional data, and ANN synapses need suitable weight values for generating a minimum error. In the last stage of this study, we performed an early intervention/engagement ex periment on those learners who showed weak performance in their study. The early intervention experiment’s objective was to encourage and help vulnerable learners in crease their learning behavior by sending motivational and adaptive triggers/messages on their mobile devices. For the early intervention experiment, 425 learners were se lected (learners having D and F grades). They were further divided into the control group and experimental group learners (the control group had 212 learners, whereas the experimental group received 213 learners). Two types of triggers were sent on ex perimental group learners, namely adaptive and motivational triggers. In contrast, the control group learners were independent of receiving any adaptive content and moti vational/adaptive triggers. The early intervention experiment lasted for 50 days. Its result concluded that intervened learners showed overall better performance than un intervened learners in the final learning activity (Intervened learners had overall 12.30 percent higher performance than un-intervened learners). Lastly, we used the End User Computing Satisfaction (EUCS) model questionnaire to measure learners’ atti tude towards using an intelligent learning system. EUCS questionnaire was shared only iii with 213 experimental group learners as the control group was independent of using an intelligent learning system during the early intervention experiment. Six dimensions of client-side application, namely learnit, were used to determine learners’ satisfac tion level towards it. The six dimensions of EUCS included usefulness, ease-of-use, engagement, timeliness, adaptiveness, and attitude towards learning software applica tions. Overall, the scores recorded for all six dimensions of EUCS were greater than 4, which inferred that overall, the experimental group learners were satisfied with in telligent learning system functionalities and will use such software applications in the future.
Gov't Doc #: 23558
Appears in Collections:PhD Thesis of All Public / Private Sector Universities / DAIs.

Files in This Item:
File Description SizeFormat 
Muhammad Adnan CS 2021 kust kohat.pdfPhd. Thesis5.4 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.